Sonia Pujol, Ph. D Brigham and Women's Hospital, Harvard Medical School Boston, MA
Slicer Ribeirão Preto Workshop June 30, 2025
Medical images have traditionally been manually segmented, which is a time-consuming process that requires intensive effort by radiologists and is subject to inter-reader variability.
Text BodyIn the past decade, image segmentation has been powered by the development of deep learning algorithms (e.g. nnUnet by the German Cancer Research Center (DKFZ)/Helmholtz Research).
AI-powered segmentation tools can reduce the segmentation time and provide more reproducible results.
A is an AI algorithm that was trained to perform a specific task (e.g. brain tumor segmentation model).
The Weights of an AI model are small numbers that determine how much importance the model gives to different image features.
During the Training phase, a Model learns patterns from data labelled by experts and adjusts its weights to improve its predictions.
During the Validation/Test phase, the model is evaluated on a separate set of data not used during the Training phase.
During Inference, the model is applied to new datasets to perform the specific task it was trained for.
This tutorial focuses on running inference tasks using various pretrained AI models for automated segmentation of anatomical and pathological structures.
This tutorial uses the pre-trained models of the MONAIAuto3DSeg Slicer extension.
The tool is designed to work on laptops or on average desktop computer without a GPU.
Text BodyMultiple modalities Support (CT, MRI).
Multiple anatomies (head, thorax, abdomen, pelvis, etc.).
Multiple pathologies (tumor, hemorrhage, edema).
Segmentation Task #1: Prostate.
Segmentation Task #2: Brain Glioma.
Segmentation Task #3: Whole Body Segmentation.
Text BodyAI-based Segmentation of Peripheral Zone (PZ) and Transition Zone (TZ) of the prostate on T2-weighted MRI Images.
Dataset:
msd_prostate_01-t2
msd_prostate_01-adc
AI-based Segmentation of Neoplasm, Necrosis and Edema in Brain MRI images.
Datasets:
1) BraTS-GLI_00005-000-t1n (T1-weighted)
2) BraTS-GLI_00005-000-t1c (T1-weighted post-Gd)
3) BraTS-GLI_00005-000-t2w (T2-weighted)
4) BraTS-GLI_00005-000-t2f (T2-FLAIR )
AI-based Segmentation of the whole body.
Dataset:
CT_ThoraxAbdomen
The 3D SlicerMONAIAuto3DSeg extension provides fast AI-based segmentation of anatomical and pathological structures.
The module can run on standard laptop and desktop computers with no GPU.
The 3D Slicer internationalization project and the 3D Slicer for Latin America project have been made possible by two CZI Essential Open Source Software for Science (EOSS cycle 4 & 5) grants.